Sentinel-2 and Landsat Derived Suspended Sediment Concentrations

Abstract

The dynamics of river suspended sediment, derived from soil erosion, is critical for understanding floodplain and coastal wetland evolution, as well as reservoir sedimentation. Although the U.S. Geological Survey (USGS) has collected > 105 suspended sediment concentration (SSC) samples, data availability is often sparse or altogether lacking for large river transects. Landsat derived SSC measurements have proven accurate enough to supplement USGS datasets, allowing unprecedented spatial analysis of SSC trends throughout large river systems (Dethier et al. 2020). Here, I build on this approach by applying it to higher spatial and temporal resolution datasets. I have derived suspended sediment concentrations from the Sentinel-2 satellite sensor through a cluster and regression approach. To increase the number of training samples, I constructed SSC-discharge rating curves for all in-situ USGS stations. This has constrained the uncertainty of Sentinel-2 derived SSC to less than a factor of two, which has proven adequate for large rivers. In combination with the Landsat record, this allows for a multi-decadal analysis of sediment transport dynamics across multi-dammed systems. This study applies these methods to the Chattahoochee River in Georgia and Alabama, USA. Using observations from 1984 to 2022, there exists pronounced decreases in SSC downstream of dams along the river, with downstream reaches never regaining the same values as upstream observations. Also evident is a decreasing trend in SSC temporally, which could be indicative of changing land-use practices. Code for this project is publicly available at github.com/ivalencius/sentinel-ssc.